On 3/14/26 12:39 AM, Paul Richard Thomas wrote:
Hi Jerry,
Yes, I agree that it is absurd but also understandable. I have kept
clear of both OMP and coarray bugs, especially regressions, because I
don't feel competent to deal with them.
I suggest the following policy in respect of partially implemented
regression fixes:
1] Fixes already in place on 26th April 2023 (the gcc-13.0 release)
should be closed;
This makes sense.
2] If possible backport regressions to gcc-13 and close, if no more
than pattern matching prevents application of the patch and regtesting
is OK;
Good
3] Do not backport if other fixes or evolutionary changes prevent it; and
A real possibility, I can easily check these.
4] In cases 1] and 2], an informatory message should be sent to the
fortran list. Flagging several closed PRs at the same time is
acceptable. In the case of 3], an explanatory message to the fortran
list of the intention to close, with a deadline.
This also makes sense to me. I don't want to perturb the process too much. I am
thinking two 'business' days after notification here. (Keeping in mind most
people should have their weekends off)
Thanks Paul, and others feel free to weigh in.
Jerry
Of the regressions assigned to me:
110626: I do not see a satisfactory way to fix this bug without
reworking ordinary assignment to be fully compliant with F2018 10.2.1;
84245: A patch was posted on 2024-11-30 but a query on the PR
prevented further progress. I will return to it and, if I cannot see a
way to prevent the fix from causing a compiler memory leak, will
submit it to the list;
10155: A patch was posted on 2025-02-08 that "needs to be completely
refactored". It still applies cleanly and runs OK; and
105168: A patch was posted on 2025-01-03. It applies cleanly and gives
the same runtime results as at least one other brand. (2 others ICE!)
Paul
On Sat, 14 Mar 2026 at 02:20, Jerry D <[email protected]> wrote:
On 3/13/26 7:13 PM, Jerry D wrote:
Hello all,
During Chris Alberts campaign going through all the gfortran know regressions he
noted a large number that have been fixed and never backported. These simpley
hang out there to bug and distract us and nothing every gets done.
Most of these are either openmp or oacc bugs. Myself I see no point in back
porting these. Others think we should backport. My goal is zero regressions.
If the consensus is to backport these, then my first response is the initial
patch committer should have done it by now. Some of these are many years old.
I will backport these and test, but I do not do the usual other libgomp tests
that are run by others, so I hesitate for fear of breaking something since some
of the patches are ancient.
The question is backport and close or just close them? If backport, how far back
is worth going, obviously 13 is as far back as we can go? (Since no one has
whined about these regressions for so long my impression is they are irrelevant)
Regards,
Jerry, master janitor extraordinaire.
PS This is a classic example, this thing has sat out in never never land for 10
years. This is absurd.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77371
